AGENTLAS MODEL 01 · THE ENGINE INSIDE AGENTLAS
Model codename
Not a prompt generator. The engine that builds and runs your agents.
Hephaestus is the engine inside Agentlas. Give it one idea and it builds a single agent, a team of agents, or a clean package you can share. New skills only stick after they've proven they work — so your agents get better without going off the rails.
Hephaestus isn't about making more agents. It's about making agents you can actually trust and keep.
“The best agent builder isn't more magic. It's giving you real control over what your agents do.”HOW THE ENGINE WORKS
Hephaestus runs every request through four stages — intake, mode routing, governed memory, and portable runtime output — so agent creation and packaging become a repeatable system instead of a one-off prompt.
Hephaestus doesn't rush your one-line idea straight into files. First it works out what the agent is for, where it will run, what it's allowed to touch, and what should stay private — so what gets built is actually what you meant.
Before building anything, Hephaestus decides which of three things you actually need — a single agent, a team of agents that work together, or a clean package you can share — so one prompt never gets mashed into a confusing all-in-one.
When an agent picks up a new skill, Hephaestus doesn't trust it right away. The skill has to show it actually works first. Only then does the agent start relying on it — so a bad habit can't quietly take over.
The same agent works in Claude Code, Codex, Gemini CLI, Agentlas Desktop, and AGENTS.md. The real instructions live in one open place, so you're never locked into a single app.
WHAT'S INSIDE
Reads what you asked for and decides whether to build one agent, a team, or a package
Actually makes the single agent, the team, or the cleaned-up package
One open place (the .agentlas folder) that holds what the agent remembers, how it's organized, and the rules it follows
Make the same agent run in Claude Code, Codex, Gemini CLI, Desktop, and AGENTS.md
Run quality, safety, and install tests before anything ships
WHY IT MATTERS
Every package comes with a plain README and a clear folder layout, so you can tell what each part does at a glance.
It runs the same way in Claude Code, Codex, Gemini, and Agentlas Desktop — no separate hidden version for each one.
Built-in checks confirm the agent actually installs and that nothing private leaks before you publish it.
New skills and habits only stick after they've proven they work, so the agent improves over time without drifting off the rails.
VISION
Agentlas isn't only about automating tasks. It keeps your agents, what they remember, and their safety checks in plain, open structures you can look at, change, and trust — on GitHub and on your own computer.
See the public code on GitHub
